A systematic review linking welfare and automated analysis of acoustic signals in broiler chickens and layers

SUMMARYAutomated vocalisation analysis is advancing to link chicken behaviour with welfare assessment. This systematic review investigates which vocalisations produced by broilers and laying hens can be detected through sound analysis and how they relate to animal welfare, using the Five Domains model and the PICo framework. The search strategy ((broiler OR chicken) AND (sound OR vocalisation OR ‘voice recognition’)) was applied across PubMed, Scopus, Web of Science, and SciELO, identifying 57 relevant peer-reviewed articles published between 2000 and 2025. Seventeen distinct vocalisation types or noises were identified. Most studies were conducted under experimental conditions and focused on the on-farm production phase. In the nutrition domain, pecking sounds showed high predictive accuracy for feed intake and growth, but indicators like hunger, thirst, or malnutrition remain underexplored. In the physical environment domain, vocalisations signalled thermal discomfort and poor air quality, though responses to stocking density, ammonia, and other factors need further study. The health domain was most researched, linking coughs and sneezes to diseases, though most models were developed in small, controlled settings. Behavioural interactions were least studied, mainly focusing on feather pecking in laying hens. The mental state domain links vocalisations to emotions, but challenges remain in interpreting their consistency and contextual meaning. Only 36% of studies met preferred recording standards, with most relying on basic statistical methods over advanced machine learning. Model performance often dropped in real-world settings, highlighting the need for standardised protocols and robust, generalisable training data. Much more work is needed to understand what vocalisations mean, how they vary by context, and how reliably they reflect welfare. No studies addressed transport or slaughter phases. Future research should prioritise standardisation, open data sharing, and broader coverage across all production phases to support practical and meaningful welfare monitoring applications.